PupRN: um método para diagnóstico de anormalidades oftalmológicas em recém-nascido baseado na dinâmica pupilar

• Ano de defesa: 2021
• Autor(a) principal: Silva, Marcos Vinicius Ribeiro
• Orientador(a): Camilo Júnior, Celso Gonçalves
• Banca de defesa: Camilo Júnior, Celso Gonçalves, Naves, Eduardo Lázaro Martins, Rosa, Thierson Couto
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal de Goiás
• Programa de Pós-Graduação: Programa de Pós-graduação em Ciência da Computação (INF)
• Departamento: Instituto de Informática - INF (RG)
• País: Brasil
• Palavras-chave em Português: Anormalidade oftalmológica; Dinâmica pupilar; Recém-nascido; Sistema automatizado de pupilometria; Diagnóstico
• Palavras-chave em Inglês: Ophthalmological abnormality; Pupil dynamics; Newborn; Automated pupillometry system; Diagnosis
• Área do conhecimento CNPq: CIENCIAS EXATAS E DA TERRA::CIENCIA DA COMPUTACAO
• Link de acesso: http://repositorio.bc.ufg.br/tede/handle/tede/11733

Vision is one of the human senses that help development since birth, being of paramount importance for cognitive, social, and motor skills. The World Health Organization (WHO) points out that the number of children with ophthalmic abnormalities should increase by about 200 million between 2000 and 2050. Dynamic pupillometry is an exam that captures immutable pupillary behavior, such as its change in involuntary size, aiming to diagnose eye disorders and diseases. Since these pathologies being severe in children and the potential of pupillometry analysis for their diagnosis, this work proposes a method for diagnosing ophthalmic abnormalities using machine learning techniques and intelligent algorithms. Thus, the method autonomously extracts pupillary information from pupillometry exams and applies a classifier model to distinguish newborns between normal and altered clinical conditions within the ophthalmological context. This model intends to be a trial screening method that could help health professionals diagnose newborns' ophthalmological abnormalities. In addition, an annotated benchmark, which was manually developed in this study, is available and presents the context and highlights the obstacles in working with pupillometry exams in newborns. The algorithms proposed by this work were evaluated and compared with the ElSe and ExCuSe algorithms, state-of-the-art algorithms in the subject of pupillary tracking applied to the scope of this study. In conclusion, it presented a classifier model capable of differentiating newborns with diseased diagnosis in the ophthalmic field with an accuracy close to 81% under the available dataset.